TWI614721B - Detection of defects embedded in noise for inspection in semiconductor manufacturing - Google Patents

Abstract

One embodiment relates to an apparatus for detecting defects on a manufactured substrate. The apparatus includes an imaging tool configured to obtain an image frame from the fabricated substrate. The apparatus further includes a data processing system including a computer-configurable group configured to calculate features for pixels in an image frame and to divide the pixels in the image frame into pixels Read the code. The computer readable code is further configured to select a feature defining group and to generate a multidimensional feature distribution for the selected feature defining group. Another embodiment relates to a method for detecting defects from a test image frame and a plurality of reference image frames. Other embodiments, aspects, and features are also disclosed.

Description

Detection of defects embedded in noise used in semiconductor manufacturing

The present invention generally relates to wafer and reticle inspection devices and methods of using wafer and reticle inspection devices.

Automated inspection and inspection systems are important in program control and yield management for the semiconductor and related microelectronics industries. These systems include optical and electron beam (e-beam) based systems.

In the manufacture of semiconductor devices, early defect detection in development and manufacturing processes has become increasingly important to shorten product development cycles and increase manufacturing throughput. Advanced wafer and reticle inspection systems are used to detect, view and classify defects and return root cause information back to the manufacturing process to prevent such defects from continuing. The size of the associated defect is directly proportional to the design rules applied to the fabrication of the semiconductor device. As the design rules of the application continue to shrink, the performance requirements on the inspection system increase in both imaging resolution and rate (defects per hour of processing).

An embodiment relates to a method for detecting defects from a test image frame and a plurality of reference image frames. The features of the pixels in the test image frame and the reference image frame are calculated, and the pixels in the image frame are divided into feature defining groups of pixels. A feature defining group is selected and a multi-dimensional feature distribution is generated for defining the group for the selected feature.

In addition, one of the multi-dimensional feature distributions can be determined to be normal clusters, and abnormal points outside the normal cluster can be detected. Defective pixels associated with anomalous points can be located and defective pixels can be marked and/or reported.

One embodiment relates to a device for detecting defects on a manufactured substrate. The apparatus includes an imaging tool configured to obtain an image frame from the fabricated substrate. The apparatus further includes a data processing system including computer readable code configured to perform one of the methods of detecting defects, the method dividing pixels into feature defining groups and generating group-specific feature distributions Detect abnormal points.

Other embodiments, aspects, and features are also disclosed.

200‧‧‧Image frame/frame area

202‧‧‧ vertical line

204‧‧‧ horizontal line segment

206‧‧‧Open space

300‧‧‧Characteristic distribution graph

302‧‧‧Normal cluster

304‧‧ points

400‧‧‧Characteristic distribution graph

402‧‧‧Normal cluster

404‧‧‧ points

510‧‧‧ imaging tools

520‧‧‧Active substrate support

525‧‧‧Target substrate

530‧‧‧Detector

540‧‧‧Data Processing System

542‧‧‧ processor

544‧‧‧ memory

545‧‧‧ computer readable code

546‧‧‧Data Storage System

550‧‧‧System Controller

552‧‧‧ processor

554‧‧‧ memory

555‧‧‧Computer readable code

556‧‧‧Data Storage System

1 is a flow chart depicting one method of detecting defects in image data in accordance with an embodiment of the present invention.

2 is a diagram depicting one of an exemplary detection region within an image frame in accordance with an embodiment of the present invention.

3 and 4 show an exemplary two-dimensional feature map generated from an image frame of one of the detection regions similar to the framed region depicted in FIG.

Figure 5 is a schematic illustration of one of the detection devices that can be utilized for automated inspection of a fabricated substrate in accordance with an embodiment of the present invention.

According to an embodiment of the invention, images from at least two dies may be used to perform defect detection. An image associated with a die (where the defect is to be detected) may be referred to as a test image, and an image associated with other die may be referred to as a reference image.

This application discloses an innovative technique to improve the sensitivity of defect detection during the inspection of fabricated semiconductor substrates. In particular, the technical improvement detection can be sensitive to the defects of "embedded in noise" in a feature distribution graph (multidimensional signal) on a multidimensional feature space.

To form a feature distribution graph, points in the multidimensional feature space may be assigned a population value. For example, if there are one hundred pixels in the frame area with the same test and reference features, the population value of the feature points is assigned to one hundred.

In the technique previously disclosed in U.S. Patent No. 7,440,607, since the feature profile is derived from a test image and a plurality of reference images, the feature profile is preferably formed based on features of pixels within the area of a frame. Since multiple eigenvalues of a qualified (normal) pixel fall within a particular nominal range, a normal cluster (normal distribution) is typically formed by the feature distribution of the qualified pixels in the feature distribution graph. A defective pixel becomes an abnormal point in the feature distribution curve because one or more of its characteristic values do not fall within the nominal range.

To improve defect detection, more features can be used to generate a feature profile. However, Applicants have determined that when more than two dimensions are used, the normal cluster (normal distribution) formed by the qualified pixels in the feature profile often becomes sparse, making it difficult to distinguish anomalies from normal clusters.

In contrast, in accordance with an embodiment of the present invention, a pixel of a frame is divided into individual feature definition groups based on one or more reference features. Subsequently, one of the groups of pixels can be selected, and a feature profile can be formed based on the test and reference features of the pixels within the selected group (excluding pixels outside the group).

Applicants have determined that this technique is advantageous in that it allows for the detection of previously undetected defects. This is because the previously undetected defects are embedded in the "noise" around the normal cluster of all pixels based on a frame. In contrast, when a normal cluster defines pixels in a group based on a selected feature, the defects become detectable anomalies. This is because when a different distribution graph is created for a different pixel group, the noise that can be attributed to the non-selected group can be effectively removed from the graph.

1 is a flow chart depicting one method 100 of detecting defects in an image data in accordance with an embodiment of the present invention. The image data may include a test image and a corresponding reference image. Reference and test images can be divided into several frames. Each frame can cover a part of a die A significant number of pixels in the area. The frame can be rectangular or square. For example, 512 x 512 or 1024 x 1024 pixels can form a square frame.

Feature calculation

The first step 102 of the method 100 involves calculating reference features and test features of pixels in the reference and test images. Reference and test features can be determined as follows.

A reference feature can be defined as some of the properties associated with the same (corresponding) pixel location on one of the plurality of reference images. For example, the reference feature can be the average or median of the gray levels across multiple reference dies at the same pixel location. As another example, a reference feature can be a range or deviation of gray levels of a plurality of reference dies at pixel locations.

One of the reference features originating from a pixel location may also include information around the pixel location. For example, a local value range or a local average of one of three pixel wide by three pixel high regions (or other local regions defined relative to pixel locations) centered at a pixel location on each reference image may be first calculated, and then The local value range or average can be used to derive a reference feature for the pixel location. For example, a median value that is one of the local averages across multiple dies may be one of the reference features of the pixel location. Many other features can be derived from these principles.

A test feature can be derived from a test image and a pixel location on a plurality of reference images. An example of a test feature is the difference between the grayscale on the test image and the average of the grayscales on the reference image. Another example of a test feature is the difference between the grayscale on the test image and the midtone of the grayscale on the reference image. Similar to a reference feature, one of the test features originating from a pixel location may also contain information around the pixel location. For example, a local average around one of the three-pixel wide by three-pixel high regions (or other local regions defined relative to the pixel locations) at the pixel location on the test image can be used to calculate for multiple reference images. The difference between the average values of the local averages around the three-pixel wide by three-pixel high region of one of the corresponding positions.

Feature definition group and group specific feature distribution

In a second step 104, the pixels of a frame can be divided into feature defining groups. in In this step, at least one feature is used to separate at least one group of pixels from other pixels in the detection region. For example, a pixel of a densely patterned region typically has local grayscale value domain variations for pixels in more than one open region. Therefore, a local gray scale value field can be used as one feature for dividing pixels into different group groups.

In a third step 106, a multi-dimensional feature distribution can be generated for one or more selected groups of pixels. In contrast, a prior art would use all of the pixels of an image frame to produce a multi-dimensional feature distribution to be used for abnormal point detection.

For example, consider the detection area as shown in image frame 200 depicted in FIG. As shown, image frame 200 includes vertical line 202, horizontal line segment 204, and open space 206.

A prior art would use all of the pixels in frame 200 to produce a multi-dimensional feature distribution. In contrast, the present disclosure technique first divides pixels in the frame 200 into groups based on calculating features in accordance with one of the second steps 104.

In a first example, using a local grayscale range or other computational feature, the pixels of frame 200 can be divided into a first group comprising one of the pixels having both vertical line 202 and horizontal line 204 and including having an open A second group of pixels of space 206. In a second example, the pixels of frame 200 can be divided into a first group comprising one of the pixels having a vertical line 202, a second group comprising one of the horizontal line segments 204, and a pixel having an open space 206. A third group.

Subsequently, one or more of the feature defining groups may be selected according to a third step 106, and a multi-dimensional feature distribution may be generated independently for each of the selected groups.

Normal clustering decision

A fourth step 108 of method 100 may determine pixels that fall within a normal cluster (normal distribution) of points for each group-specific feature distribution. In accordance with embodiments of the present invention, there may be several effective methods for defining a normal cluster of pixels.

One method for identifying normal clusters is based on populations in local neighborhoods of one of the point locations of the signal distribution formed in the multidimensional feature space. For 2D instances, A predetermined population density threshold may be used for determining whether a given point is normal for one of a total population of points within a one-to-five-pixel wide by five-pixel high square region centered at a given point in the two-dimensional signal distribution. One of the limits. If the population value is greater than the population density threshold, the point can be considered normal.

Another method for identifying normal distribution is based on connectivity among the points in the signal distribution. For a two-dimensional instance, if there are other points within a predetermined distance in the signal distribution, then one point can be considered normal.

Other methods can also be used to define the normal distribution. As an example, if one point satisfies the above two criteria, the point can be considered normal. In principle, if there are a significant number of pixels with the same or similar characteristics, they are considered normal and have no defects.

Abnormal point detection

The fifth step 110 can be identified as a pixel of the statistical anomaly point, which can be an indication of a defect on the test substrate. As described above, a normal cluster of points in a group-specific feature distribution has been identified. Points that are not identified as being within a normal cluster can be considered a candidate point. A candidate point can contain one or more pixels corresponding to one or more actual defects.

To allow for some margin of error, allow a tolerance range before declaring a defect. Different methods can exist in defining the tolerance range. One method is a fixed tolerance range. In a two-dimensional example, the tolerance range may be a predetermined distance from a normal point to one of the candidate points. If the candidate point is placed at a distance from any normal point that is one distance away from the predetermined distance, it can be declared as a defective point because the pixel associated with the candidate point can be considered defective. More complex rules, such as making the predetermined distance a function of one of the reference features, may also be added to the defined tolerance range. As an example, the distance may be a function of one of the reference features defined as one of the average gray levels across the plurality of reference dies.

To achieve the best performance in identifying actual defects and eliminating process nuisance, parameters such as tolerance ranges are tuned. Each defect point can be viewed in the multi-dimensional feature space, and the tolerance range can be tuned to obtain a point corresponding to the defect of interest, and the corresponding system is excluded. The point of the defect.

Defective pixel recognition

As described above, anomalous points may be detected in a particular set of features during a fifth step 110, and pixels associated with the outliers may be considered defective. It should be noted that in a multi-dimensional feature space, each point represents one or more pixels having a particular characteristic characteristic. However, the actual pixel location at which the pixel is in the frame region may not remain directly in the feature distribution. Therefore, additional processing must be performed to locate such defective pixels in accordance with the sixth step 112.

Once the defective pixels associated with the abnormal points are located, the defective pixels can be marked and reported according to the seventh step 114. In one embodiment, the tool operator and/or an engineering database can be reported so that the defect can be studied. In some embodiments, the means in which the method of detecting the defect is implemented may also include methods for detecting analysis and identification. In other embodiments, the two functions of detection and analysis can be implemented separately.

Instance feature distribution curve

3 and 4 illustrate an exemplary two-dimensional image generated from an image frame of one of the detection regions of the frame region 200 (which has a vertical line 202, a horizontal line segment 204, and an open space 206) similar to that depicted in FIG. Feature distribution graph. One of the darker points in the graph indicates a larger population of pixels having feature values within the range of values associated with the point.

The feature distribution graph 300 in FIG. 3 is generated from all of the pixels in the image frame 200. The normal cluster 302 can be clearly seen in the graph. In this case, there is a point 304 associated with the actual defective pixel. However, in this feature distribution 300, the point 304 falls in the "noise" near the edge of the normal cluster 302. Therefore, the point 304 is not recognized as an abnormal point in this distribution.

In contrast, the feature distribution graph 400 in FIG. 4 is only generated from the pixels of the vertical line 202 in the image frame 200. Normal cluster 402 is also clearly visible in the graph. The normal cluster 402 in FIG. 4 is substantially smaller (closer) than the normal cluster 302 in FIG. In this case, the point 404 associated with the pixel of the actual defect is outside of the normal cluster 402. So here In the feature distribution 400, the point 404 is identified as an abnormal point.

High level diagram of the device

Figure 5 is a schematic illustration of one of the detection devices that can be utilized to fabricate a substrate in accordance with an embodiment of the present invention. As shown in FIG. 5, the detecting device includes an imaging tool 510, a movable substrate support 520, a detector 530, a data processing system 540, and a system controller 550.

In an embodiment, imaging tool 510 includes an electron beam (e-beam) imaging line. In an alternate embodiment, imaging tool 510 includes an optical imaging device. In accordance with an embodiment of the present invention, imaging tool 510 includes electronics to control and adjust the magnification of the imaging.

The movable substrate support frame 520 can include a translation mechanism to hold a target substrate 525. The target substrate 525 can be, for example, a semiconductor wafer or a reticle for lithography. The detector 530 is a suitable detector for one of the particular imaging devices, and the data processing system 540 is configured to process the image data from the detector 530. The data processing system 540 can also include a processor 542, a memory 544 for storing computer readable code 545, a data storage system 546 for storing data, and various other components, such as a system bus, input/output. Interface and so on.

System controller 550 can be communicatively coupled to imaging tool 510 to electrically control the operation of imaging tool 510. The system controller 550 can also include a processor 522, a memory 554 for storing computer readable code 555, a data storage system 556 for storing data, and various other components, such as a system bus, input/output. Interface and so on.

in conclusion

In the above description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the invention. However, the above description of the present invention is not intended to be exhaustive or to limit the invention to the precise form disclosed. Those skilled in the art will appreciate that the present invention may be practiced without one or more of the specific details or other methods, components and the like.

In other instances, structures or operations that are well known are not shown or described in detail to avoid obscuring aspects of the invention. While the invention has been described with respect to the specific embodiments of the present invention and the embodiments of the present invention, it will be understood by those skilled in the art that various equivalent modifications may be within the scope of the invention.

Such modifications of the invention are possible in light of the above detailed description. The terms used in the following claims are not to be construed as limiting the invention. In fact, the scope of the invention should be determined by the following claims, which should be construed in accordance with the principles defined in the claims.

510‧‧‧ imaging tools

520‧‧‧Active substrate support

525‧‧‧Target substrate

530‧‧‧Detector

540‧‧‧Data Processing System

542‧‧‧ processor

544‧‧‧ memory

545‧‧‧ computer readable code

546‧‧‧Data Storage System

550‧‧‧System Controller

552‧‧‧ processor

554‧‧‧ memory

555‧‧‧Computer readable code

556‧‧‧Data Storage System

Claims (16)

An apparatus for detecting a defect on a manufacturing substrate, the apparatus comprising: an imaging tool configured to obtain an image frame from the manufacturing substrate; and a data processing system including a processor, a memory, and Computer readable code in the memory, the computer readable code configured to calculate features for pixels in an image frame; the value of the features having values within a plurality of specific ranges The pixels in the image frame are separated from other pixels in the image frame such that the separated pixels form one of the pixel feature defining groups; only the image frame that defines the group for the feature belonging to the pixel The pixels generate a multi-dimensional feature distribution; determine a normal cluster in the multi-dimensional feature distribution; and detect anomalies in the multi-dimensional feature distribution outside the normal cluster. The device of claim 1, wherein the computer readable code is further configured to locate defective pixels associated with the abnormal points. The device of claim 2, wherein the computer readable code is further configured to report the defective pixels. The device of claim 1, wherein the features comprise reference features, and wherein one of the reference features is associated with one of pixel locations located on the plurality of reference images. The device of claim 4, wherein the features further comprise test features, and wherein one of the test features is derived from a test image and a pixel location on the plurality of reference images. The device of claim 4, wherein the property comprises a range of gray levels at the pixel location. The device of claim 4, wherein the property comprises from a center at the pixel location A partial range of pixel information. A method for detecting a defect from a test image frame and a plurality of reference image frames, the method comprising: imaging a portion of a substrate on a movable substrate support frame by an image device to generate the test An image frame; and a data processing system including a processor, a memory, and a computer readable code in the memory to perform the step, the method comprising: the test image frame and the plurality of reference image frames a plurality of pixel computing features; separating pixels in the image frame having values of the features within a plurality of specific ranges from other pixels in the image frame such that the separated pixels Forming a feature defining group of pixels; generating a multidimensional feature distribution only for the pixels belonging to the feature defining group of pixels; determining a normal cluster in the multidimensional feature distribution; and detecting outside the normal cluster Anomalous points in the multidimensional feature distribution. The method of claim 8, further comprising: locating defective pixels associated with the abnormal points. The method of claim 9, further comprising: marking the defective pixels. The method of claim 8, wherein the features comprise reference features, and wherein one of the reference features is one of a property associated with a pixel phase location on the plurality of reference image frames. The method of claim 11, wherein the features further comprise test features, and wherein one of the test features is derived from a test image frame and a pixel location on the plurality of reference image frames. The method of claim 11, wherein the property comprises a range of gray levels at the pixel location. The method of claim 11, wherein the property comprises information from a local range of pixels centered at the pixel location. A non-transitory tangible data storage medium storing computer readable code configured to perform a method, the method comprising: calculating features for pixels in a test image frame and a reference image frame; using a test image for the test image The frames and the pixels calculated by the pixels in the plurality of reference image frames are separated by separating pixels belonging to one of the feature definition groups of the pixel from other pixels of the feature definition group not belonging to the pixel; The feature of the pixel defines the pixels of the group to produce a multi-dimensional feature distribution; determining one of the normal clusters of the multi-dimensional feature distribution; and detecting anomalies in the multi-dimensional feature distribution outside the normal cluster. The non-transitory tangible data storage medium of claim 15, wherein the method further comprises: locating defective pixels associated with the abnormal points; and marking the defective pixels.